Thomas D Swinburne, CNRS

CV  /  Google Scholar  /  GitHub  /  thomas.swinburne-at-cnrs.fr

Recent technical advances
POPSRegression (ML:S&T 2025 / code):
Fast mispecification/model-form UQ (applied to SNAP/MACE)
DescriptorDOS (arXiv 2025 / code) :
Differentiable / model-agnostic free energies
ImplicitDiff (NPJ 2025 / code):
Differentiable / model-agnostic minimisation in JaX/LAMMPS

New approaches for computational materials science
Metallic alloys are uniquely strong, ductile and recyclable, but predicting how components fail remains a grand engineering challenge, limiting performance and increasing costs/emissions.

Our research pushes the boundaries of theory and simulation for materials, connecting atomic dynamics to engineering properties. Drawing broadly from mechanics, physics, statistics and machine learning, a particular focus is quantifing uncertainty in multi-scale, data-driven models.

One recent theme is model-agnostic simulations: to quantify multi-scale uncertainty and incorporate multi-scale data, simulation outputs need to be smooth functions of model parameters, not static numbers. Whilst well known in many fields (e.g. adjoint/tangent schemes in FEM) we are developing new model-agnostic approaches for atomic material simulations, for inverse fine-tuning in back-propagation and model-form UQ in forward propagation.
Keywords: dislocations, multi-scale UQ, atomic simulation, forecasting

Recent papers/preprints from the group
Differentiating through free energies (with Marinica Group) ArXiv 2025
Propagating POPS-UQ to SNAP/MACE (with Perez Group) ArXiv 2025
Implicit differentiation in MD (Ivan Maliyov) NPJ Comp. Mat. 2025 / code
POPS misspecification uncertainty ML:S&T 2025 / code
Coarse-graining & forecasting MD with descriptors Phys. Rev. Lett. 2023
Embedded ab initio with QM/ML (Petr Grigorev) Acta Mat. 2023 / code

Recent papers/preprints with collaborators
Entropy of dislocation glide (Rodney Group / PAFI) to appear Nat. Comm.
Efficient ML/ML in LAMMPS (Fraser Birks / Kermode Group) ArXiv 2025
Free energies from mean-field bonds (Raynol Dsouza) Phys. Rev. B 2024
MACE foundation model (Csyani Group, Section A.13) ArXiv 2023
A15 defects in FCC (Marinica Group / TAMMBER) Nat. Comm. 2023
Ill-conditioned Markov chains (Wales Group / PyGT) Proc. Roy. Soc. 2023

Recent & Upcoming Invited Conferences/Seminars
Dislocations 2025, Miami, USA, 11/25
CECAM: UQ from DFT to ML, EPFL, Lausanne, 11/25
CoMPASs: Math and Comp. Mat. Sci., ICMS, Edinburgh, 11/25
Speaker & Senior Scholar, IPAM, UCLA, 9/25-12/25
E-MRS Meeting, Prague, 09/25 (unavailable)
Fritz Haber Institute of the Max Planck Society, Berlin, 04/25
APS March Meeting, Anaheim, 03/25 (unavailable)
Condensed Matter Physics Seminar, École Polytechnique, 03/25
CECAM Promises of the multiscale, EPFL, Lausanne, 03/25
Keynote, MecaNano symposium, online, 02/25
AppliedML 2025, EPFL, Lausanne, 02/25